The statistical un-likelihood of Donald Trump [analysis of a Trump-Up analysis]

Eric Schulman and Daniel Debowy write:

Contrary to what you might have heard recently (Get ready for President Trump, says election whiz who’s scary accurate) from a statistics-driven political scientist, Donald Trump will not “almost certainly become President of the United States” if he wins the Republican nomination. The mainstream media are helping spread that message — while ignoring the work of actual scientists who have examined U.S. Presidential electoral data since 1789 and can tell you the honest truth: Donald Trump would almost certainly lose to Hillary Clinton.

Norpoth-and-trumpThat Trump-Up statistical adventure

Professor Helmut Norpoth‘s Primary Model has been receiving notice in the popular press for its prediction that if Donald Trump becomes the Republican nominee for President, he will have a 97% chance of defeating Hillary Clinton or a 99% chance of defeating Bernie Sanders in the upcoming U.S. Presidential Election (for example, A Statistician With a Near-Perfect Election Formula Says Prepare for President Trump).

Our better statistical adventure

As the developers of the Annals of Improbable Research Presidential Election Algorithm (“An Algorithm for Determining the Winners of U.S. Presidential Elections,”), we wish to take this opportunity to make the following observations:

1) Our algorithm agrees that Donald Trump would likely defeat Bernie Sanders (for example, it predicts that a Republican ticket of Donald Trump/Ted Cruz would defeat a Democratic ticket of Bernie Sanders/Elizabeth Warren) but predicts that Donald Trump would be almost certain to lose to Hillary Clinton (for example, Trump/Cruz would lose to a Democratic ticket of Hillary Clinton/Julián Castro).

2) Although Professor Helmut’s model correctly determines the winner of the popular vote in 25 of the 26 U.S. Presidential elections since 1912, our algorithm correctly determines the winner of all 57 U.S. Presidential elections since 1789 (including the elections of 1824, 1876, 1888, and 2000, in which the winner of the popular vote did not receive the most electoral college votes and therefore did not win the election).

3) Professor Helmut’s model ignores vice presidential candidates, whereas we believe they can have a significant impact on the results. For example, if Bernie Sanders becomes the Democratic candidate for President and chooses Lincoln Chafee as the Democratic candidate for Vice President, our algorithm predicts this ticket would defeat Trump/Cruz (although it predicts Sanders/Chafee would lose to a Republican ticket of Donald Trump/Chris Christie). Few Vice Presidential choices would allow Trump to outpace Clinton and she could pick a strong Vice Presidential choice herself, which would make it almost impossible for Trump to win.

4) Finally, whereas Professor Helmut has a Ph.D. from the University of Michigan in Political Science, one of us (ES) has a Ph.D. from the University of Michigan in actual science (Astronomy) and the other (DD) has a Ph.D. from NYU School of Medicine in Neuroscience. The U.S. Presidential Election in November 2016 may well provide a test for which of the models is correct and therefore which fields of study are best suited to predicting the results of U.S. presidential elections.